Lsqfit

==========================
Modified so that it works with numpy>=2.0.

==========================
Same as v13.2.1.
- Fixes bug in lsqfit's GSL module that prevented compilation on some systems.

- Fixes minor bug in vegas_fit affecting parameter nitn.

=============================
- Adds vegas_fit.sample, which samples the PDF used in the fit. This is useful for
making probability density histograms and contour plots.

- Minor bug fix: vegas_fit wasn't always passing correct arguments to PDFIntegrator.

=============================
Adds a second least-squares fitting strategy. lsqfit.vegas_fit uses PDFIntegrator
from the vegas Python module to evaluate means, covariances, PDF histograms, etc
using Bayesian integrals, as opposed to minimization as in lsqfit.nonlinear_fit.
Fits from nonlinear_fit are Gaussian approximations to the results from
vegas_fit.

=============================
- New fitter lsqfit.vegas_fit that uses Bayesian integration rather than minimization.

- Small changes to tests and examples to account for new formatting of GVars (v12.0).

- More robust (and simpler) implementation of lsqfit.__version__ --- no longer
uses importlib.metadata, which is buggy.

- Documentation on how to compile with the GSL library (ie, don't use the lsqfit wheels).

===========================
- New installation code so that it works on Python3.11 even when numpy
is not installed, and on systems with old versions of numpy.

- Adds Python wheels to PyPi distribution.


Version 13 2022-11-01
======================
Minor changes except that pickling works better now. This
means that multiple processors can be used when doing
Bayesian integrals using the vegas module's PDFIntegrator
(ie, with nproc>1) and the fit's PDF fit.pdf(p). This is useful
because fit.pdf(p) can be quite costly to evaluate when fits
are complicated.

- fit.pdf(p) is now normalized so that fit.pdf(fit.pmean)=1.
This differs from previous versions. To get the old
normalization use fit.pdf(p)/exp(fit.pdf.lognorm). This
change was made because exp(fit.pdf.lognorm) can easily
overflow when fits involve lots of parameters.

- New function fit.dchi2(p) replaces fit.evalchi2(p) and
fit.logpdf(p), which are deprecated.

- empbayes_fit has a new keyword p0 which specifies the
fit-parameter starting point for the first fit; p0
is set automatically in subsequent fits to optimize
fitting (unless it is specified by fitargs(z)).

- Fix to show_plots() to accommodate a change in Matplotlib.

- Forces Cython to regenerate *.c files when using Python 3.11 or later (deals
with incompatibilities introduced by 3.11).